Master of Science Thesis
Stockholm, Sweden 2016
Examensarbete
Stockholm, Sverige 2016
Ellen Cronqvist Fredrik Smed
Examensarbete INDEK 2016:117 KTH Industriell teknik och management
Industriell ekonomi och organisation
SE-100 44 STOCKHOLM
Ellen Cronqvist Fredrik Smed
Master of Science Thesis INDEK 2016:117 KTH Industrial Engineering and Management
Industrial Management
SE-100 44 STOCKHOLM
Examensarbete INDEK 2016:117
Affärsmodeller på den svenska bankmarknaden
Ellen Cronqvist Fredrik Smed
Godkänt 2016-06-14
Examinator Tomas Sörensson
Handledare
Gustav Martinsson Uppdragsgivare
Finansinspektionen
Kontaktperson Gunnar Dahlfors
Sammanfattning
Den senaste finanskrisen har visat att det finns ett behov av ökad övervakning av aktörerna på den finansiella marknaden. Ett sätt att förbättra övervakningen är genom att öka förståelsen för
företagens affärsmodeller. Syftet med detta examensarbete är att hitta likheter i affärsmodellerna hos svenska kreditinstitut och hos svenska filialer av utländska banker. Mer specifikt syftar denna studie till att hitta grupper av företag, i denna rapport kallat kluster, med liknande affärsmodell och till att identifiera existerande affärsmodeller på den svenska bankmarknaden. Informationen som användes i studien är från årsredovisningar som rapporterades till Finansinspektionen för åren 2000, 2005, 2010 och 2013.
För att möjliggöra en jämförelse mellan olika aktörers data har kvoter skapats utifrån deras balans- och resultaträkningar. För att reducera mängden data och för att få ett fåtal okorrelerade variabler användes principalkomponentanalys. Metoden som användes för att hitta klustren är en hierarkisk agglomerativ metod kallad Wards metod. Antalet kluster bestämdes genom att använda Calinski- Harabasz-index. Bootstrapping användes för att testa stabiliteten hos de identifierade klustren.
Denna studie visar att mönster existerar på den svenska bankmarknaden och att det är möjligt att hitta kluster av företag med liknande affärsmodell. Svenska filialer av utländska banker och svenska kreditinstitut har studerats separat. För svenska kreditinstitut hittades sex kluster och för att beskriva affärsmodellerna kallas de: Universalbanker, Sparbanker, Leasingföretag, Icke inlåningsfinansierade
kreditinstitut, Servicefokuserade kreditinstitut och Övriga kreditinstitut. De mest stabila klustren, det vill säga de med högst likhet, är Sparbanker och Leasingföretag. Klustret med lägst likhet är Universalbanker och detta bör ses som ett mönster i använd data snarare än ett kluster. För de svenska filialerna av utländska banker hittades tre kluster och dessa kallas: Banker, Servicefokuserade kreditinstitut och Övriga kreditinstitut. Dessa kluster är stabila.
Nyckelord
Bankers affärsmodeller, klusteranalys, bankövervakning, kreditinstitut, svenska banker
Master of Science Thesis INDEK 2016:117
Business Models in the Swedish Banking Market
Ellen Cronqvist Fredrik Smed
Approved 2016-06-14
Examiner
Tomas Sörensson
Supervisor
Gustav Martinsson Commissioner
Finansinspektionen
Contact person Gunnar Dahlfors
Abstract
The recent financial crisis has emphasized the need for improved supervision of the actors on the financial market. One way to improve supervision is through better understanding of business models. The aim with this thesis is to find similarities in business models for Swedish credit
institutions and for Swedish branches of foreign banks. More specific this study aims to find groups of companies, in this paper called clusters, with similar business models and identify existing
business models in the banking market. The data used in this study are financial statements reported to the Swedish Financial Supervisory Authority for the years 2000, 2005, 2010 and 2013.
In order to compare the companies’ data, ratios from the income statements and balance sheets have been created. To reduce the amount of data and arrive at a smaller set of uncorrelated variables, principal component analysis was used. The method used for finding the clusters was a hierarchical agglomerative clustering method called Ward’s method. The number of clusters was determined using Calinski-Harabasz index. Bootstrapping was used in order to test cluster stability.
This study shows that patterns in the Swedish banking sector exist and that it is possible to find clusters of companies with similar business models. Swedish branches of foreign banks have been treated separately from Swedish credit institutions. For Swedish credit institutions a division into six clusters was found to be most suitable and in order to describe the business model the clusters are named: Universal banks, Savings banks, Leasing companies, Non-deposit funded credit institutions, Service-focused credit institutions and Other credit institutions. The most stable clusters, that are the clusters with highest similarity, are Savings banks and Leasing companies. The cluster with lowest stability is Universal banks and it could be considered as a pattern in the data rather than a cluster. For Swedish branches of foreign banks, three clusters were found to be most suitable and the clusters are named: Banks, Service-focused credit institutions and Other credit institutions. These clusters are stable.
Keywords
Bank business models, cluster analysis, banking supervision, credit institutions, Swedish banks
1. Introduction ... 1
1.1 Background ... 1
1.2 Problem Definition ... 2
1.3 Purpose... 2
1.4 Research Questions ... 3
1.5 Delimitations ... 3
1.6 Academic Contribution... 3
1.7 Disposition ... 3
2. Institutional Background ... 4
2.1 Swedish Banks ... 4
2.1.1 Commercial Banks ... 4
2.1.2 Savings Banks ... 5
2.1.3 Members Banks ... 5
2.1.4 Foreign Branches of Swedish Chartered Banks ... 5
2.1.5 Foreign Banks with Operation in Sweden ... 5
2.1.6 Market Share of the Ten Largest Banks ... 6
2.1.7 Number of Banks over Time ... 6
2.2 Swedish Credit Market Companies ... 7
2.2.1 Credit Market Companies ... 7
2.2.2 Foreign Branches of Swedish Loan Companies ... 7
2.2.3 Foreign Credit Market Companies ... 7
2.2.4 The Largest Credit Market Companies ... 8
2.2.5 Number of Credit market Companies over Time ... 8
2.3 How Banks Generate their Revenue ... 9
2.3.1 Net Interest Income ... 9
2.3.2 Net Commission Income ... 10
2.3.3 Other Operating Income ... 10
2.4 Supervision of Swedish Credit Institutions... 10
3. Literature Review ... 12
3.1 Business Models ... 12
3.2 Strategic Groups ... 13
3.3 Business Models and Bank Stability ... 13
3.4 Business Models and Bank Risk ... 13
3.5 Better Understanding at a System Level... 14
3.6 Finding Groups with High Similarity ... 14
3.7 Interaction between Banking Sector and the Real Sector ... 15
3.8 Results in Earlier Studies ... 15
4. Methodology ... 17
4.1 Identification of Variables ... 17
4.2 Removal of Outliers ... 18
4.3 Number of Dimensions ... 18
4.4 Principal Component Analysis ... 19
4.5 Number of Principal Components ... 20
4.6 Clustering ... 20
4.7 Determining the Number of Clusters ... 21
4.8 Cluster Stability ... 22
5. The Data Set ... 24
5.1 Identification of Variables ... 24
5.2 Boxplots for Swedish Credit Institutions ... 25
5.3 Boxplots for Swedish Branches of Foreign Banks ... 26
6. Result ... 28
6.1 Swedish Credit Institutions ... 28
6.1.1 Number of Principal Components ... 28
6.1.2 Number of Clusters ... 29
6.1.3 Clustering Results ... 29
6.1.4 Cluster Stability ... 34
6.1.5 Time Perspective ... 35
6.2 Swedish Branches of Foreign Banks ... 36
6.2.1 Number of Principal Components ... 36
6.2.2 Number of Clusters ... 37
6.2.3 Clustering Results ... 37
6.2.4 Cluster Stability ... 39
6.2.5 Time Perspective ... 40
7. Discussion ... 41
8. Conclusion ... 44
8.1 Future Studies ... 44
Table 1 Total assets for the ten largest banks in 2014 ... 6
Table 2 Number of banks supervised by the Swedish FSA ... 7
Table 3 Total assets of the Swedish mortgage institutions in 2014 ... 8
Table 4 Total assets of the other credit market companies in 2014 ... 8
Table 5 Number of credit market companies supervised by the Swedish FSA ... 9
Table 6 Income for Swedish banks 2015 ... 9
Table 7 Results in earlier studies ... 16
Table 8 How the variables influence the principal components ... 28
Table 9 Calinski-Harabasz index values for different numbers of clusters ... 29
Table 10 Number of observations ... 29
Table 11 Characteristics over time for Universal banks ... 31
Table 12 Characteristics over time for Savings banks... 32
Table 13 Characteristics over time for Leasing companies ... 32
Table 14 Characteristics over time for Non-deposit funded credit institutions ... 33
Table 15 Characteristics over time for Service-focused credit institutions ... 33
Table 16 Characteristics over time for Other credit institutions ... 34
Table 17 Jaccard coefficient for the six clusters ... 34
Table 18 Number of companies in each cluster over time ... 35
Table 19 Number of companies changing to each cluster ... 35
Table 20 How the variables influence the principal components ... 36
Table 21 Calinski-Harabasz index values for different numbers of clusters ... 37
Table 22 Number of observations ... 37
Table 23 Jaccard coefficient for the three clusters ... 39
Table 24 Number of companies in each cluster over time ... 40
Figure 1 Components of a business model ... 12
Figure 2 The first two principal components span the plan that best fits the data ... 19
Figure 3 Boxplots for Swedish credit institutions ... 25
Figure 4 Boxplots for Swedish credit institutions ... 26
Figure 5 Boxplots for Swedish branches of foreign banks ... 27
Figure 6 Boxplots for Swedish branches of foreign banks ... 27
Figure 7 Average asset side of the balance sheets for all clusters... 30
Figure 8 Average liability side of the balance sheets for all clusters ... 30
Figure 9 Average income statement for all clusters ... 31
Figure 10 Average asset side of the balance sheets for all clusters ... 38
Figure 11 Average liability side of the balance sheets for all clusters ... 38
Figure 12 Average income statement for all clusters ... 38
We would like to thank Gunnar Dahlfors and Jesper Bruzelius at Finansinspektionen for introducing us to this topic and for their feedback. Without them this thesis would not have been possible. Furthermore, we would like to thank our supervisor Gustav Martinsson, Associate Professor at Royal Institute of Technology, for his guidance during the process.
Stockholm, June 2016
Ellen Cronqvist Fredrik Smed
1
This section gives an introduction to the topic and provides an understanding of the problem addressed in the thesis. The purpose is then described, research questions are given, the delimitation and academic contribution is presented and finally the disposition of the thesis is given.
The companies in the financial sector offer services which are important for economic growth and for a working economy. Difficulties for financial companies can affect the whole economy and this can lead to difficulties for non-financial companies to get access to credit and thereby limit their possibility to invest and expand (Sveriges Riksbank, 2015). A great financial crisis occurred 2007- 2008 and in order to prevent future financial crises, monitoring of credit institutions has been intensified and requirements have been raised. Credit institutions are companies that have the right to accept deposits from the public and offer transfer of payments. Credit institutions consist of banks and credit market companies. Credit institutions in Sweden are supervised by Finansinspektionen, which is the Swedish Financial Supervisory Authority (FSA). The role of the Swedish FSA is to supervise financial markets and make sure that companies on this market follow appropriate rules, in order to ensure financial stability. At the moment the Swedish FSA supervises more than 200 Swedish credit institutions (Finansinspektionen, 2016a).
Why banks exist can be explained by the economic functions they perform. One economic function
offered by banks is to connect borrowers and lenders. Another economic function of banks is
financial asset transformation, which means that individual lenders may offer different asset
quantities than the borrowers require. Banks also perform the function of screening potential
borrowers and monitoring borrowers over the period they borrow. Another important function is
offering efficient ways of transferring funds (Cavelaars & Passenier, 2012). Banks may become
bankrupt due to the nature of the business offered and there are two main reasons for this to
happen: insolvency and/or liquidity crisis. Both are often caused by counterparty risks. If asset
values decline sharply the capital of a bank can rapidly be reduced or even be wiped out. Even the
risk of this happening can reduce counterparties’ willingness to lend, which is very problematic in
situations where banks need cash or liquid assets to meet their commitments. In situations where
banks have bad assets, for example non-performing loans on their balance sheet, it takes time to
discover the true value of them and they may take many years to mature. Asset values that have
declined below previously reported values are written down and can therefore lead to insolvency
(Blundell-Wignall, et al., 2014).
2
A central part of bank supervision is to reduce financial system vulnerability and limit the companies’ risk taking. In order to prevent bank failures, regulatory frameworks and requirements are put in place. To ensure that financial activities are handled with proper care and good risk management, the Swedish FSA has put up minimum requirements. Credit institutions need to have sufficient capital to cover both known risks and risks that are not fully known. Sufficient liquidity is needed in order to cover short-term payment commitments (Finansinspektionen, 2016b).
Between 2013 and 2019 a new type of regulatory framework called Basel III is gradually implemented. It was designed to improve the ability of banks to absorb losses and decrease the probability for new crises. The European Union has decided to implement Basel III regulations in its member states and the rules should govern all banks and credit market companies (European Union, 2013).
The European Banking Authority has prepared guidelines in order to achieve common procedures and methodologies for supervision of banks. The framework for Supervisory Review and Evaluation Process is based on four pillars and one is business model analysis (European Banking Authority, 2014). The Swedish FSA also points out that there is a need to understand companies’ business models in order to prioritize the supervision (Finansinspektionen, 2016b). Analyzing bank business models is the focus of this thesis and more specifically it investigates whether it is possible to divide Swedish credit institutions into clusters with similar business models.
Currently there are more than 200 credit institutions in Sweden and it is time consuming for the Swedish FSA to monitor them individually. For the Swedish FSA, it is relevant to examine whether companies have similar business models. If they do it opens up a possibility to simplify the supervision, since it might be possible to supervise them in clusters rather than individually. By studying groups of companies, characteristics over time can be found. If a company deviates from previous year’s group characteristics it can act as an indication for the Swedish FSA to investigate whether there has been significant changes in the operations of the company.
The purpose of this study is to find similarities in business models for Swedish credit institutions
and for Swedish branches of foreign bank. This study aims at finding clusters of companies with
similar business models and to identify the existing business models for credit institutions. It also
aims at investigating whether companies belong to the same category over time or if they move
between clusters.
3
This thesis aims at answering two main questions where one of them can be divided into two sub- questions. The research questions are:
How should Swedish credit institutions be categorized?
This question can be divided into the following sub-questions:
o How many categories are suitable?
o Which are the distinct features for each category, in terms of how companies in each category generate their income and how their balance sheets are built up?
Do Swedish credit institutions change categories over time?
This study classifies Swedish credit institutions and Swedish branches of foreign banks. Foreign branches of Swedish banks are not studied separately, since they are included in the Swedish banking groups. Only active companies were considered for this study, in other words, companies with discontinued operations were not included. This study uses data for the years 2000, 2005, 2010 and 2013.
As opposed to earlier studies, this study includes credit market companies and all small banks, while earlier studies have mainly focused on large banks. The inclusion of credit market companies differentiates this study from earlier studies, although it is important to bear in mind that some of these companies could have been classified as banks in other jurisdictions. No previous study has, to our knowledge, categorized Swedish credit institutions and Swedish branches of foreign banks, which our study aims to do. In previous studies the authors have selected ratios they found useful for characterizing business models. Our ambition is to use another method where ratios do not need to be selected manually.
The report is structured as follows. Chapter 2 describes the institutional background. Chapter 3 describes a literature review within this area and how this study differs from previous studies.
Chapter 4 describes the methodology. Chapter 5 presents the data and describes it by using
boxplots. Chapter 6 presents the identified clusters both for Swedish credit institutions and for
Swedish branches of foreign banks. Chapter 7 answers the research questions and discusses the
results. Chapter 8 concludes the study and presents suggestions for future research.
4
This section describes the institutional background of this study. Initially, the Swedish credit institution market is described and afterwards an introduction to supervision of Swedish credit institutions is given.
At the moment the Swedish FSA supervise 165 Swedish banks and 520 foreign banks with operation in Sweden (Finansinspektionen, 2016a). These banks can be divided into groups according to the laws the companies are obliged to follow and the groups are described below.
At the moment there are 39 banking companies (limited liability companies) (Finansinspektionen, 2016a). Below follows a description of the three types these companies can be divided into.
The Swedish bank market is dominated by four large banks and these are: Skandinaviska Enskilda Banken AB, Nordea Bank AB, Swedbank AB and Svenska Handelsbanken AB. These banks offer a wide range of services and can be considered as universal banks (Svenska Bankföreningen, 2016). In Sweden, these four banks had 74 % of total bank assets in 2014. Among the four large banks, Nordea can be differentiated by having a larger share of lending to foreign public, approximately ¾ of their public lending is to foreign public (Sveriges Riksbank, 2015).
The parent companies for the four large banks own several companies where different segments of the business are conducted. For example the mortgage operation of Svenska Handelsbanken AB is in the company Stadshypotek AB and the leasing business is in Handelsbanken Finans AB. Svenska Handelsbanken AB also has separate companies for insurance and mutual funds. Stadshypotek AB and Handelsbanken Finans AB are credit market companies, which mean that they are supervised separately.
Due to the financial crisis in Sweden in the early 1990s, it became possible to restructure savings
banks into banking companies. In 1992, 11 restructured savings banks merged and formed
Sparbanken Sverige, which was listed on the Swedish stock exchange in 1995. In 1997 Sparbanken
Sverige merged with Föreningssparbanken and 2006 the name was changed to Swedbank AB. When
a savings bank is restructured into a banking company, an ownership foundation is created
5
(Sparbanksakademin, 2016). A restructured savings bank can be partly or fully owned by the ownership foundation (Sparbankernas Riksförbund, 2013). At the moment there are 13 restructured savings banks, some of these are partly owned by Swedbank AB.
Several banks specialized in different areas exist on the Swedish banking market. Many of these can be considered as niche banks, which mean that they are not offering all the services that are offered by some of the larger banks. These banks are mainly concentrated in the private customer market and they often do not operate with physical branch offices. Instead the services are offered online or by telephone (Svenska Bankföreningen, 2016). One example is Avanza, an online stock broker.
At the moment there are 47 savings banks in Sweden (Finansinspektionen, 2016a). Savings banks operate in a local market and are in most cases small. A savings bank has no shareholders and therefore do not pay any dividends. Instead, the profit is kept in the bank (Sveriges Riksbank, 2015), but usually a part of the profit is donated to local associations, for example sports clubs (Sparbankernas Riksförbund, 2013). The number of savings banks has decreased over time, mostly due to mergers and conversion into banking companies. For example: Lönneberga sparbank merged with Lönneberga-Tuna-Vena sparbank in 2008 and Sparbanken Alingsås, Sparbanken Eken and Sparbanken Skaraborg converted to banking companies.
A members bank is an economic association offering bank services to its members. Only two companies are classified as members banks in Sweden: JAK Medlemsbank and Ekobanken medlemsbank (Finansinspektionen, 2016a). Members banks have a market share of less than 0.1 % in Sweden.
At the moment the Swedish FSA supervise 77 foreign branches of Swedish chartered banks (Finansinspektionen, 2016a). This group consists of foreign branches of Swedish banking companies, usually operating in European countries and especially in the Nordic countries. The majority of foreign branches of Swedish chartered banks belong to the four largest Swedish banks.
The presence of foreign banks in Sweden is significant and foreign banks operate in Sweden as branches. A branch is not a separate legal entity; instead it is a part of the foreign company.
Branches do not have any share capital, the assets and liabilities belong to the foreign company.
However, a branch must have its own accounting, which is separated from the foreign company
(Bolagsverket, 2015). At the moment, there are 28 foreign management companies (branches)
(Finansinspektionen, 2016a). The largest one is Danske Bank, which is the fifth largest bank in
6
Sweden in terms of total assets (Sveriges Riksbank, 2015). Danske Bank has had a great presence in Sweden since its acquisition of the Swedish bank Östgöta Enskilda Bank in 1997.
The Swedish banking market is characterized by a few large banks and a large number of small banks. Table 1 shows total assets of the ten largest banks in 2014. The four largest banks, in terms of total assets are Skandinaviska Enskilda Banken AB, Nordea Bank AB, Swedbank AB and Svenska Handelsbanken AB. In addition to the four largest banks there are a number of other large banks:
Danske Bank A/S, SBAB Bank AB, Länsförsäkringar Bank Aktiebolag, DNB Bank ASA, Landshypotek Bank Aktiebolag and Skandiabanken Aktiebolag (Sveriges Riksbank, 2015). Danske Bank A/S and DNB Bank ASA are foreign branches, while the others are banking companies (Finansinspektionen, 2016a).
Bank Total assets 2014
(Billion SEK) Skandinaviska Enskilda Banken AB 1,601
Nordea Bank AB 1,590
Swedbank AB 1,214
Svenska Handelsbanken AB 1,026
Danske Bank A/S 861
SBAB Bank AB 157
Länsförsäkringar Bank Aktiebolag 126
DNB Bank ASA 95
Landshypotek Bank Aktiebolag 82
Skandiabanken Aktiebolag 51
Sum 6,803
Table 1 Total assets for the ten largest banks in 2014 (Sveriges Riksbank, 2015)
The sum of total assets for the ten largest banks in Sweden in 2014 was 6,803 billion SEK and the sum of total assets for all banks in Sweden was 7,371 billion SEK (Sveriges Riksbank, 2015). This illustrates that the Swedish banking market is dominated by a few large banks and the majority of banks have very small market shares.
The number of banks supervised by the Swedish FSA has changed over time which can be seen in
Table 2. The total number of Swedish banks has increased from 137 in 2000 to 165 in 2016. The
number of banking companies has increased from 22 to 39 and the number of foreign branches of
Swedish chartered banks has increased from 29 to 77. The number of members banks has been
constant over the period and the number of savings banks has decreased from 84 to 47. During the
same period the number of foreign management companies has increased from 19 to 28
(Finansinspektionen, 2016a; Finansinspektionen, 2016c).
7
2000 i 2005 2010 2013 2016
Banking companies (limited liability company) 22 29 36 39 39
Foreign branches of Swedish chartered banks 29 39 57 65 77
Members banks 2 2 2 2 2
Savings banks 84 76 53 49 47
Swedish banks, total 137 146 148 155 165
Foreign management company (branch) 19 19 26 27 28
Table 2 Number of banks supervised by the Swedish FSA
A credit market company is a limited liability company authorized to conduct financing (SFS 2004:297). A credit market company can both engage in borrowing and lending. Historically, banks had a monopoly in deposits, but since July 2004 credit market companies are allowed to offer credit and receive deposits (Sveriges Riksbank, 2015). At the moment the Swedish FSA supervises 78 credit market companies and 32 foreign credit market companies (Finansinspektionen, 2016a).
These companies can be divided into groups according to the laws applicable to the companies. The groups are described below.
At the moment the Swedish FSA supervises 36 credit market companies (Finansinspektionen, 2016a). A wide range of services are offered by credit market companies and typically these companies are focused on offering credit in a particular area, for example mortgages. Sveriges Riksbank (2015) divides the credit market companies into two groups: mortgage institutions and other credit institutions. Mortgage institutions mainly finance home loans and at present there are six companies. The category other credit market companies consists of finance companies and corporate- and municipality-financing institutions. The majority of other credit market companies are finance companies and they usually focus on different types of funding, for example leasing or factoring (Sveriges Riksbank, 2015).
At present, 42 foreign branches of Swedish loan companies are supervised by the Swedish FSA (Finansinspektionen, 2016a). This group consists of foreign branches of credit market companies which usually operate in European countries and especially in the Nordic countries.
In order to do business in Sweden, foreign credit market companies need permission from the Swedish FSA. At the moment there are two Swedish branches of foreign credit market companies (Finansinspektionen, 2016a).
i 2000-2013 as 1st January. 2016 as May 2016
8
In 2014, total assets of all credit market companies was 3,598 billion SEK of which mortgage institutions accounted for 2,693 billion SEK. This can be compared to total assets for all banks, branches included, that was 7,371 billion SEK at the same time (Sveriges Riksbank, 2015). It is important to remember that most of the larger credit market companies are owned by banks. Table 3 describes total assets of the Swedish mortgage institutions in 2014.
Mortgage institution Total assets 2014
(Billion SEK)
Swedbank Hypotek AB 917
Stadshypotek AB 908
Nordea Hypotek Aktiebolag 491
AB Sveriges Säkerställda Obligationer 229
Länsförsäkringar Hypotek AB 148
Sum 2,693
Table 3 Total assets of the Swedish mortgage institutions in 2014 (Sveriges Riksbank, 2015)
Table 4 describes total assets of the ten largest other credit market companies in 2014.
Other credit market companies Total assets 2014 (Billion SEK)
Kommuninvest i Sverige AB 337
Aktiebolaget Svensk Exportkredit 323 Handelsbanken Finans Aktiebolag 46
Nordea Finans Sverige AB 45
Volkswagen Finans Sverige AB 26
Wasa Kredit AB 15
Hoist Kredit Aktiebolag 14
Entercard AB 10
Toyota Material Handling Europe Rental 9
Svenska Skeppshypotekskassan 7
Sum 832
Table 4 Total assets of the other credit market companies in 2014 (Sveriges Riksbank, 2015)
Table 3 and Table 4 show that the group of credit market companies is dominated by a few large companies and the largest companies are mortgage lenders. A large number of smaller credit market companies are also operating in the market.
The number of credit market companies supervised by the Swedish FSA has changed over time
which is shown in Table 5. The total number of Swedish credit market companies has decreased
from 105 in 2000 to 78 in 2013. The number of foreign branches of Swedish loan companies has
increased from 26 to 42 and the number of credit market companies has decreased from 79 to 36.
9
During 2005 and 2010 the Swedish FSA supervised one credit market association. During the same period the number of foreign credit market companies has increased from 0 to 2 (Finansinspektionen, 2016a; Finansinspektionen, 2016c).
2000 ii 2005 2010 2013 2016
Foreign branches of Swedish loan companies 26 32 41 43 42
Credit market company 79 74 57 47 36
Credit market association 0 1 1 0 0
Credit market companies, total 105 107 99 90 78
Foreign credit market company, branch office 0 1 3 2 2
Table 5 Number of credit market companies supervised by the Swedish FSA
Nowadays banks offer more services than just lending and borrowing. As can be seen in Table 6, the main sources of income are interest income, commission income and dividends from group companies. Below the table are two main sources of income, net interest income and net commission income, explained further. Some examples of what is included in other operating income is presented afterwards.
Income for Swedish banks 2015 (banks,
savings banks and Swedish branches) Billion SEK
Interest income 113
Interest expense -46
Leasing income 22
Commission income 56
Commission expense -16
Net result on financial operations 9
Dividends received 55
Of which from group companies 53
Other operating income 15
Total income 209
Table 6 Income for Swedish banks 2015 (Statistiska Centralbyrån, 2016)
Net interest income is the difference between interest income and interest expense. Net interest income was the greatest source of income for Swedish banks in 2015. The main part of interest income is derived from loans to the public. One example of how interest income and expeneses are distributed for one of the largest banks can be found in Skandinaviska Enskilda Banken AB’s annual report for 2015 (Skandinaviska Enskilda Banken, 2016). Approximately 70 % of interest income was
ii 2000-2013 as 1st January. 2016 as May 2016
10
derived from loans to the public with an average interest rate of 2.0 %. Approximately 25 % of the interest expense was paid to public depositors with an average interest rate of 0.4 %. Around ⅔ of the interest expense was paid on issued debt securities with an average interest rate of 1.6 %. For savings banks the great majority of interest income comes from loans to the public and almost the whole interest expense is paid to deposits from the public. One example can be found in the annual report for Sörmlands Sparbank in 2015, which is one of the largest savings banks in Sweden (Sörmlands Sparbank, 2016). In this case 97 % of the interest income was from loans to the public and 84 % of the interest expense was paid to deposits from the public.
Net commission income is the difference between income and expense for fee-based services. Banks offer a wide range of fee-based services. For a large bank, significant commission generating activities are for example asset management, credit cards, payment proccessing and brokerage. One example of how commission acitivities are classified by a large bank can be found in Nordea’s annual report for 2015. In this report commission activities are classified as either savings related commission, payment commission, lending related commission and other commission income (Nordea, 2016). Some companies also offer services such as insurance or corporate finance.
Acitvities that do not fit into the other categories are categorized as other operating income.
Examples of what could be included in other operating income are divestment of shares, remunerations from group undertakings, IT services, profit from sale of properties and income from real estate operations.
The Swedish FSA’s classification of banks is made according to the applicable laws. All Swedish banks and credit market companies are under the rule of law The Banking and Finance Business Act (SFS 2004:297). This law describes general conditions the companies have to follow. There are special laws for different types of banks. Savings banks have to follow the Savings Banks Act (SFS 1987:619). Members banks have to follow the Members Banks Act (SFS 1995:1570). Foreign banks with Swedish branches have to follow the Foreign Branch Offices Act (SFS 1992:160). The Swedish FSA supervises approximately 2,000 financial companies and 900 foreign financial companies with operations in Sweden (Finansinspektionen, 2016a).
The Swedish government has ordered the Swedish FSA to be responsible for supervision and
licensing for financial markets and financial companies. In addition to the government’s decree
regarding supervision, Swedish laws, for example The Banking and Finance Business Act (SFS
2004:297) and Special Supervision of Credit Institutions and Investment Firms Act (SFS 2014:968),
control the Swedish FSA’s operations. The special Supervision of Credit Institutions and Investment
Firms Act states that the Swedish FSA is responsible for checking that companies follow prudential
11
regulation, which is a regulation stating that companies need to hold adequate capital (Finansinspektionen, 2016b).
All companies, independent of their size, have to follow the basic requirements and the supervision
is, in principle, the same. Supervised companies have to report financial data to the Swedish FSA
which is used to analyze key performance indicators as well as risk indicators. In order to prioritize
supervision an understanding of the business models is required. Knowledge about how risks
emerge and how revenue is generated can be used to prioritize the supervision. The four largest
banks are more supervised than other companies, not only because they are significantly larger than
all other financial companies but also because they have a high interconnection with the Swedish
financial system (Finansinspektionen, 2016b).
12
This section first describes the business model concept in general and then how it can be applied on banks. A literature review regarding clustering of banks is then presented according to the purpose of the study.
Figure 1 Components of a business model (Shafer, et al., 2005)
As can be seen from the previous chapter, there are significant differences between Swedish credit institutions. Both in terms of size and what they are specialized in. How a company conducts its business can be characterized by its business model. However, there is no generally accepted definition of what a business model is. One definition is that it is a representation of a company’s strategic choices and underlying core logic (Shafer, et al., 2005). Shafer et al. (2005) classify the components of a business model into four distinct categories: strategic choices, the value network, creating value and capturing value. The strategic choices consider for example which markets to target and which products to offer. Value network can be described as the context within which a firm competes and solve problems for customers (Christensen & Rosenbloom, 1995). Creating value and capturing value consider two important aspects of companies needs to be viable over a longer time period (Shafer, et al., 2005). The components are general and apply to all types of companies.
When considering banks specifically, Cavelaars and Passenier (2012) argue that a description of a
bank’s business model needs to, at least, describe what the bank offers (in terms of products and
services), how it distributes what it offers, how it reaches its potential customers, how it generates
profit and discuss if the profits are sustainable.
13
Mergaerts and Vander Vennet (2016) found the European banking sector to be characterized by a continuum rather than a discrete set of business models. In contrast, this thesis aims to find clusters that are as homogenous as possible, as done in earlier studies such as Ayadi et al. (2011), Ayadi and De Groen (2014) and Roengpitya et al. (2014). According to Mergaerts and Vander Vennet (2016) the business model of a bank could be described by a set of variables that captures the asset, liability, and capital and income structure. When considering other relevant variables such as distribution channels and type of products offered, the authors believe that this information should be reflected in the balance sheets and income statements for the companies.
The concept of strategic groups, made popular by Porter (1980), has been applied on banks in order to determine strategic groups within the banking sector. Strategic groups are often defined as groups of companies within the same industry who makes similar decisions in key areas. The concept of strategic groups was created to explain intra-industry performance differences. In a study by Amel and Rhoades (1988) of 16 selected bank markets, it was found that approximately six different strategic groups exist in banking and the strategy choices were similar across the studied markets.
The obtained results were stable over time. In contrast, José Más Ruíz (1999) studied the Spanish banking market and found significant changes in the number and strategies of the identified strategic groups over time. Koller (2001) applied the strategic group concept in a study of the Austrian banking market during 1995-2000. One finding from this study was that all banks do not belong to strategic groups.
The purpose of some studies, for example Köhler (2015), is to analyze the impact of business models on bank stability. Banks from 15 EU countries were studied and a large number of unlisted banks were included. Most of the unlisted banks were savings and cooperative banks and ⅔ of the banks in the study belonged to these categories. Four different types of business models were identified: savings, cooperative, commercial and investment banks. One of the findings was that banks will be more profitable and stable if they increase their share of non-interest income. Halaj and Zochowski (2009) used cluster analysis to identify strategic groups in the Polish bank sector in order to see how they differ in performance. By doing so the authors attempted to get a better ex ante assessment of the loss absorption capabilities for banks, which is important in the analysis of bank sector stability.
Altunbas et al. (2011) investigated the connection between bank risk variables and other variables
characterizing banks. In this study the impact of business models on bank risk was also investigated
and the authors found the relationship to be highly non-linear. Several relationships were found, for
example, market funding increase the distress probability for the riskiest banks, but have no impact
14
on the less risky ones. The authors suggest that regulators should increase their understanding of bank business models and how business models are connected to incentives for risk taking.
In a study by Ayadi et al. (2011) of 26 large European banks, three major business models were identified: retail banks, investment banks and wholesale banks. Retail banks can be characterized by the fact that customer deposits are the primary source of funding and that they offer loans to the customers. Investment banks are in this case banks that have considerable trading and derivatives activities. Wholesale banks are banks that have focus on domestic business and are active in the wholesale markets. The study also investigated how different business models relate to bank performance, risk characteristics and systemic stability.
How bank business models affect performance and risk is examined by Mergaerts and Vander Vennet (2016) in a study of 505 European banks over a period from 1998 to 2013. One of the findings was that retail-oriented banks performed better when considering both stability and profitability. Another finding was that diversification is positively correlated to profitability.
Ayadi and De Groen (2014) studied banking business models in Europe in order to get a better understanding at a system level of, for example, risk behavior. The study was focused on large and systemic banking groups, in other words, banks that might trigger a financial crisis if they go bankrupt. Four different business models were identified: investment, wholesale, diversified retail and focused retail. Diversified retail and focused retail can be distinguished by their reliance on debt markets and customer deposit respectively.
Some studies, for example Ferstl and Seres (2012), have an objective of finding sets of variables
reflecting business models of the banks and to find groups of banks with high similarity. In this
study five different clusters were identified. The authors use three different income sources, loan-to-
deposit ratio and loan-to-asset ratio to characterize the business models. Another study with a
similar objective is Roengpitya et al. (2014), but this study also focused on tracking how banks have
changed their business models over time. In the study by Roengpitya et al. (2014) three bank
business models were identified using balance sheet characteristics. The three identified business
models were: retail-funded commercial bank, wholesale-funded commercial bank and capital
markets-oriented bank. The first two business models mainly differ in how the banks are funded,
while the third differs by being more active in the interbank market.
15
In a study by Vagizova et al. (2014) of Russian credit institutions, the authors used cluster analysis for determining business models. The report focused on how credit institutions interact with the real economy, which is the part of the economy that produces goods and services. One important aspect in this case is the opportunity for long-term lending to the real sector with banking sector stability in mind. According to the authors, there is an asymmetry in the interaction between the banking sector and the real sector due to differences in profitability and the risks taken for generating investment income. A conflict of interest between the central bank and the credit institutions occurs when financial resources from the central bank, aimed for long-term support of the real sector, are used by the credit institutions for making short-term profits.
In Table 7 results from earlier studies are presented. Only studies that found distinct groups are included in the table. For example, Mergaerts and Vander Vennet (2016), who found the European banking sector to be characterized by a continuum rather than discrete groups is therefore not presented. In earlier studies, there are great differences between studied areas and number of studied banks. For example, Ayadi et al. (2011) studied 26 major European banks while Vagizova et al.
(2014) studied 836 Russian credit institutions. Roengpitya et al. (2014) and Ayadi et al. (2011)
focused on the largest banks and three general business models were identified: retail, wholesale and
investment banks. In contrast, studies limited to a single country identified a larger number of
business models, in the range of four to six business models. Some authors did not name the
identified clusters, which makes it difficult to describe them and only the identified number of
clusters for those studies are presented in Table 7.
16
Study Studied area Studied banks Results
Amel & Rhoades (1988) United States 16 urban banking markets from eight states for the years: 1978, 1981 and 1984
Six different clusters
Koller (2001) Austria 35 largest Austrian banks between
1995 and 2000 Five different clusters
Halaj & Zochowski (2009) Poland 48 Polish banks between 1997 and
2005 1. Universal banks
2. Corporate banks 3. Car finance banks 4. Mortgage banks 5. Retail banks 6. Regional banks Ayadi et al. (2011) European Union 26 major European banks between
2006 and 2009 1. Retail banks
2. Investment banks 3. Wholesale banks
Ferstl & Seres (2012) Europe 234 European banks between 2005
and 2011 Five different clusters
Ayadi & De Groen (2014) European Economic Area
(EEA)
147 large EEA banks between 2006
and 2013 1. Investment
2. Wholesale 3. Diversified retail 4. Focused retail Roengpitya et al. (2014) Global 222 banks from 34 countries
between 2005 and 2013 1. Retail-funded
2. Wholesale-funded 3. Trading
Vagizova et al. (2014) Russia 836 Russian credit institutions in
2012 11 clusters grouped into four
business models
Table 7 Results in earlier studies
17
This section describes the methodology used in this study. Clustering of data can be done in many different ways. The applied methods vary greatly depending on which type of data to be clustered and the purpose of the clustering. The main focus in this study is extracting the most significant patterns in the data set in order to capture differences between Swedish credit institutions. The purpose is not to classify every company into a specific category – the goal is to find which general business models that exist. Swedish credit institutions and Swedish branches of foreign banks are clustered separately, since branches are part of larger foreign companies.
In earlier studies, the variables used for clustering are selected by the authors based on their own expert or analytical judgment. This selection is subjective by nature and the selected variables are different from study to study. With no consensus regarding variables selection and with the belief that we would not do a good selection that would characterize companies active in the Swedish banking market; a different method has been applied in order to avoid this subjectivity. Instead of selecting variables, the used method creates new variables from a large set of original variables that captures the most essential information in the data set. In our opinion, the greatest drawback with the used method is that the individual variables might have great variance. By selecting a few variables for clustering, the observations in a cluster are more likely to have homogenous values for the variables used for clustering. Another great difference from earlier studies is that this study analyses if the found clusters actually are real clusters by considering the cluster stability.
The methodology can be described as follows. First variables are identified from the data set and then observations considered as outliers are removed. Afterwards, observations considered as outliers are removed and the number of variables in the data set are then reduced from a high number of correlated variables to a lower number of uncorrelated ones using principal component analysis. This step includes deciding a suitable number of variables. Then hierarchical clustering is done on the observations using the lower number of variables. A suitable number of clusters is then decided and checked for stability. These steps are described in detail in the following sections. All programming was done in the programming language R.
The data used in this study are the annual financial data reported by companies to the Swedish FSA.
Both balance sheet and income statement information are included. Data are expressed in Swedish
Krona (SEK) and in order to obtain comparability between companies of different size, variables are
formed as ratios. The denominator is either expressed as total assets, total equity and liabilities or
18
total income depending on the numerator. The posts in the reported data with very low impact, in other words, very low values in the numerator compared to the denominator, are aggregated in order to remove variables that have very low impact on the results. For example, prepayments and accrued income are included in other income instead of having them as separate variables. The posts used for forming variables are the posts in the standard report, used by the Swedish FSA for reporting financial data; see Chapter 5 and Appendix A for definitions.
The results from agglomerative hierarchical clustering (which is the method used for clustering) could be spoiled by a single outlier (Hennig, 2007). One definition of an outlier is an observation which appears to be inconsistent with the remainder of the data set (Barnett & Lewis, 1994).
Clusters consisting of a single or a few observations do not make up general business models and are therefore of very limited interest. The reason for companies to become extremely different from the others is in most cases because they are part of a larger corporate group. In this case outliers would also influence the dimensionality reduction since outliers increase variance and influence how the principal components are found.
If the asset side does not equal the liability side or the income statement variables does not equal total income, the observation is classified as an outlier, also observations that have one or several of the income statement variables above 1.5 or below -0.5 are considered as outliers. A value of 1.5 means that for example net interest income is 150 % of total income and therefore at least one other income variable must have a negative value in order for the sum to be 100 %. Approximately 1 % of the observations could be considered as outliers. If observations with extreme values in one or several variables would be included, they would most likely yield clusters with one or a few observations due to how hierarchical clustering algorithms are designed. The selected range is chosen manually but the observations that are considered as outliers have extreme values, for example 3 or 4. Extreme observations of this magnitude would most likely form a separate cluster.
The data set contains 19 variables formed as balance sheet and income statement variables. Every variable in the data set can be seen as a dimension, for example an observation could be described as a point in a 19-dimensional hyperspace if the number of variables is 19. A small number of dimensions are relevant in most cases. The dimensions that are irrelevant could produce noise and also mask the real clusters to be discovered (Dash, et al., 2010). In order to get a meaningful clustering analysis with the most relevant information, dimensions are reduced before clustering the data.
Dimensionality reduction is a concept of transforming high-dimensional data into a meaningful
representation with reduced dimensionality. One type of dimensionality reduction is feature
reduction, which is about extracting features by projection of high-dimensional data into a lower-
19
dimensional space by using algebraic transformation. By applying a concept called principal component analysis to the data set the number of dimensions is reduced. At the same time, correlated variables are transformed to a reduced set of uncorrelated ones (Dash, et al., 2010).
As mentioned in the previous section, a concept called principal component analysis (PCA) is applied in order to reduce the number of dimensions for a data set with a large number of variables that are correlated and at the same time keep as much as possible of the variation in the data set (Jolliffe, 2002). The idea behind PCA is that all observations are in a hyperspace with several dimensions, but all these dimensions are not equally relevant. PCA finds the dimensions that are most relevant (James, et al., 2013). These dimensions are called principal components (Dash, et al., 2010). The directions of the first two principal components span the plane that minimizes the sum of squared distances from each point to the plane, in other words the plane can be seen as the one best fitting the data.
Figure 2 The first two principal components span the plan that best fits the data (James, et al., 2013)
PCA can be mathematically defined as the projection of a data matrix 𝑋 on an 𝐴-dimensional
subspace by using a projection matrix 𝑃
′where object coordinates are given by a matrix 𝑇. The
columns in matrix 𝑇 are called score vectors and the rows in 𝑃
′are called loading vectors which are
the direction coefficients of the principal component plane (Wold, et al., 1987). In Figure 2 the
loading vectors would span the illustrated plane. The score vectors can be seen as the projection of
the observations on the plane spanned by the principal components. The score vectors and the
loading vectors are orthogonal. The difference between the projections and original coordinates are
called the residuals, which are in the matrix 𝐸. In Figure 2 the residual is the difference between the
data points and their projections on the plane spanned by the two principal components. 𝑥 is the
mean vector (Wold, et al., 1987).
20
PCA in matrix form is the least square model:
𝑋 = 𝑥 + 𝑇𝑃
′+ 𝐸 (1)
A basic assumption when using PCA is that the score and loading vectors corresponding to the largest eigenvalues are containing the most useful information and that the remaining ones mainly consist of noise (Wold, et al., 1987). An eigenvalue is a scalar associated with an eigenvector, which is a non-zero vector that does not change direction when linearly transformed. A larger eigenvalue means that the principal component has greater variance, in other words it contains more information. The principal components are therefore often written in order of descending eigenvalues. For further explanation of PCA and mathematical derivations, see Jolliffe (2002).
iiiIn order to determine a suitable number of principal components, several rules and techniques exist.
In this thesis, the rule called size of variances of principal components is used for determining a suitable number of principal components. This is a rule-of thumb, mainly justified by the fact that it works in practice and is intuitive (Jolliffe, 2002). The rule is described more in detail below.
The rule used is size of variances of the principal components, also called Kaiser’s rule. It is based on the idea that if all variables of the data matrix are independent, the principal components are equal to the original variables and therefore have unit variances in the correlation matrix. Any principal component with variance less than 1 contains less information than one of the original variables and could therefore be discarded. If the data set has groups of variables with large within- group correlations, but small correlations between the groups, then one principal component associated with the group has variance above 1 and the other principal components associated with the group have variances below 1. Therefore the rule will in general retain only one principal component associated with each group of this type. The calculated matrix, is in this case, the covariance matrix. In order to apply the rule on the covariance matrix, the cut-off is set to the average value of the eigenvalues of the covariance matrix (Jolliffe, 2002).
The objective of using clustering methods is to discover groups whose members are close to each other and are well separated from the other clusters (Halkidi, et al., 2001). The clustering of data is not an exact science as there is no single solution to how data should be clustered. Classification of the same data set done by different clustering algorithms can differ significantly from each other due to the use of different clustering criteria (Gordon, 1998). If the chosen algorithm is the most optimal
iii The function used in R for calculating principal components is prcomp which uses singular value decomposition for the calculations.
21
for the data set in question is very difficult to validate. For this study a statistical clustering algorithm called Ward’s method is used. Within the field of clustering banks into different business models, the algorithm has been used in earlier studies such as Halaj and Zochowski (2009) and Roengpitya et al. (2014). The selection of the clustering algorithm is done based on an assumption that the data set in this study is similar to the data used in the mentioned earlier studies and therefore Ward’s method would be suitable also in this case. The method proposed by Ward (1963) is an agglomerative hierarchical clustering procedure, which means that each observation starts on its own and is successively built into groups. First the number of groups is equal to the number of observations. In the following steps, two groups from the previous step that minimize the loss of information are merged. This means that two observations are merged if no other pair of observations can be found that are closer to each other. Distance is calculated using Euclidean distance. The loss of information for a group of observations can be calculated by for example using the error sum of squares (ESS), which is used in this study.
It is calculated as:
𝐸𝑆𝑆 = ∑ 𝑥
𝑖2𝑛
𝑖=1
− 1
𝑛 (∑ 𝑥
𝑖)
𝑛
𝑖=1 2
